研究目的
To propose a novel approach for unsupervised change detection in remote sensing images by integrating indicator kriging with Dempster–Shafer theory to handle conflicting information and improve accuracy.
研究成果
The proposed DSK method effectively combines multiple DI features using DS theory and indicator kriging, significantly improving change detection accuracy by handling conflicting information. Experimental results on three datasets show superior performance in terms of lower overall error and higher Kappa coefficient compared to benchmark methods. The approach is robust and generalizable, with potential for extension to other remote sensing data types.
研究不足
The method relies on the selection of DI generation algorithms and parameters (e.g., Tu, Tc, R), which may need adjustment for different datasets. It is tested only on Landsat images and may require modifications for other types like high-resolution or SAR images. Computational complexity is higher than single-DI methods.
1：Experimental Design and Method Selection:
The study uses a decision-level fusion method combining indicator kriging and DS theory. It involves generating a set of four difference images (DIs) using different comparison operators (CVA, SCM, PCA, SGD) to capture magnitude, direction, and shape changes. Fuzzy logic and DS theory are applied for fusion, and indicator kriging is used for reclassification of conflicting pixels.
2：Sample Selection and Data Sources:
Three multispectral remote sensing datasets are used: Neimeng dataset (Landsat-5 TM images from 2006 and 2011), Liaoning dataset (Landsat-7 ETM+ images from 2001 and 2002), and Hunan dataset (Landsat-8 OLI images from 2013 and 2016). Reference maps are created manually with ENVI software.
3：6). Reference maps are created manually with ENVI software.
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: Remote sensing images from Landsat satellites, ENVI software for reference map creation, and computational tools for implementing algorithms (e.g., FCM clustering, DS theory, indicator kriging).
4：Experimental Procedures and Operational Workflow:
Preprocessing (radiometric and geometric correction, co-registration), DI generation, mass function determination using FCM, DS fusion, adaptive partitioning based on conflict degree, and reclassification using indicator kriging. Parameters like Tu, Tc, and kriging window radius R are tested and optimized.
5：Data Analysis Methods:
Quantitative evaluation using missed detections (MD), false alarms (FA), overall error (OE), and Kappa coefficient (KC). Statistical significance tested with McNemar's test. Computational time recorded for complexity comparison.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
